Comparative study of image outlier detection using hexagonal lattice and rectangular lattice based on self-organizing map

Abstract

From the past of the year, Mammogram has drawn an attention for the clinical analysis where it play an important roles for the extraction of information that can be use in diagnosing and treatment disease. According to the significant use of the mammogram for diagnosing, the aim of this project is to detect the outlier from the mammogram images of breast through the clustering method using one of the popular and widely use in Artificial Neural Network Clustering which is Self- Organizing Map (SOM). The outlier indicate as the observation that is far from the rest of the rest of data where it can represent that the data either the unusual data or noise that is important for the further analysis. But, before proceeding to clustering process for the outlier detection, the image must be preprocessing first because the source of the images usually provides with different size and quality where it affect the accuracy of the analysis. Image preprocessing involves the process of crop the region of interest, enhanced the image, remove the noise, and do normalization. Once the image has been preprocessing, the data of the image is extract using Nonnegative Matrix Factorization (NMF). NMF has been proven as a powerful method for non-negative data such as image and document. The data that have been extracting from the image using NMF method, the extracted data is apply to the SOM technique to cluster the similarities of the data and at the same time can detect the outlier which is refer to the data that is not in any of clustering. Two type of lattice which is rectangular and hexagonal lattice will use for the training process and compare to find the lattice that can produce the best result.